development of a procedure to make better use of …...気象衛星センター技術報告 第55号...
TRANSCRIPT
気象衛星センター技術報告 第 55 号 2011 年 2 月
– 1 –
Development of a Procedure to Make Better Use of SSM/I Data Including Navigation Errors
KUMABE Ryoji* and EGAWA Takumu**
Abstract In the course of preparing the observation data for JRA-55, some navigation errors were found in
special-sensor microwave imager (SSM/I) data obtained from Defense Meteorological Satellite Program (DMSP) satellites. We therefore developed a method of eliminating from the SSM/I TDR/SDR files any erroneous data caused by such navigation errors. Based on a very simple satellite navigation model, this method enables most of the errors to be eliminated without the need for any information other than that contained in the SSM/I data files. A description of a computer program for navigation error compensation that we developed based on this method is also provided.
1. Introduction
The special sensor microwave imager (SSM/I) installed
on Defense Meteorological Satellite Program (DMSP)
satellites is an important data source for the assimilation
of meteorological data in a long-term reanalysis project
(JRA-251) being conducted by the Japan Meteorological
Agency (JMA) that covers a 26-year period from 1979 to
2004. The JMA has just started its next reanalysis project
(JRA-55), which covers the 55-year period from 1958 to
2012. We reviewed the treatment of observation data so
as to properly assimilate as many of them into the project
as possible. SSM/I data are, of course, one of various
types of observation data that needed to be reviewed. In
the course of our review, we discovered that a small
number of SSM/I data were erroneous. The errors found
were mainly related to navigation and observation time.
We found that these could be eliminated or corrected
through the implementation of a proper quality check
(QC) procedure, and so developed a computer program
for this purpose. We are now in the process of adopting
this computer program for the JRA-55 project.
Section 2 of this paper describes the SSM/I data we
investigated, section 3 describes the problems we
identified with these data and the solutions we found to
these problems, and section 4 contains some remarks on
the characteristics and the usage of the SSM/I data.
2. SSM/I data used for the JRA-25 and JRA-55
projects
In this section, we describe the data that we used in the
earlier JRA-25 project and will be using in the ongoing
JRA-55 project. The SSM/I data used in both these
projects are basically identical. SSM/I data from DMSP-
08, DMSP-10, DMSP-11, and DMSP-13 were provided
by the National Climatic Data Center (NCDC) of the
National Oceanic and Atmospheric Administration
*Climate Prediction Division, Global Environment and Marine Department, Japan Meteorological Agency **Numerical Prediction Division, Forecast Department, Japan Meteorological Agency
(Received February 12, 2010, Accepted December 15, 2010) 1 http:// jra.kishou.go.jp/
METEOROLOGICAL SATELLITE CENTER TECHNICAL NOTE, No.55, FEBRUARY, 2011
– 2 –
(NOAA) on magnetic tapes when we started the JRA-25
project. Covering the period from July 1987 to 1997, the
data provided on these tapes was in the antenna
brightness temperature data record (TDR) format.
Unfortunately, we were unable to extract a few of the data
from the tapes. We have downloaded post-1998 SSM/I
data from NOAA’s web data services, obtaining historical
data from NOAA’s Comprehensive Large Array-data
Stewardship System (CLASS2) and real-time data from
its National Weather Service Telecommunication
Gateway (NWSTG).
The quality of the SSM/I data was mainly checked by
manual observation for the JRA-25 project, and this
manual QC revealed that a few of the data had navigation
errors and were poorly calibrated. We were aware that a
few erroneous data that escaped detection in the manual
QC had been included in the project’s assimilated data,
but felt that the effect that such errors had on the
reanalyzed meteorological data could be ignored.
For the JRA-55 project, we review the SSM/I data
stored at the JMA to enable us to provide better
observation data. NCDC kindly provided us with a
complete inventory for the period from 1994 to 2008 of
the SSM/I data stored in their CLASS archive. Using this
inventory, we were able to obtain all the data that had
been unavailable to us for the JRA-25 project. CLASS’s
data cover the period from July 1994 to the present date
(as of December 2009, the range of data stored on
CLASS had been extended to cover the period up to
December 1990). As the pre-1997 SSM/I data on CLASS
are in netCDF format, the pre-1997 data held at the JMA
that were downloaded after the completion of the JRA-25
project are also in this format.
We now download SSM/I data from the NWSTG on a
daily basis. These data are then compared with the SSM/I
data inventory on CLASS every month. If our
comparison reveals that the inventory contains some data
that we had not obtained through our daily downloads, we
download the missing data from CLASS to ensure that
we have all the available data. We have been able to
acquire almost identical data as those in the CLASS
archive by following this careful acquisition process.
Inventories of SSM/I data possessed by the JMA are
provided in Excel files named “inventory_fNN.xls”
(where “NN” represents the number of the DMSP
satellite from which the data were obtained) together with
information regarding any erroneous data which we
detected in a QC.
3. Problems with and solutions to the collection of
SSM/I data
While decoding and preprocessing the SSM/I data, we
found a few problems with regard to data quality. Some
of these problems have arisen quite regularly since
satellite observations were begun. There were too many
erroneous data for the problem to be ignored or for a
manual check and correction to be performed. We
therefore developed a method of identifying navigation
errors in the data, and then created a computer program
for navigation error compensation based on this method.
We have not developed a QC algorithm for brightness
temperature, but presume that such data will be treated in
the QC process of the assimilation. The computer
program corrects erroneous data where possible and
discards them if correction is impossible. The SSM/I data
that is output by the program will be assimilated in the
JRA-55 project.
The remainder of this section describes the navigation
errors found in the SSM/I data, as well as the method
used to correct/discard such errors.
3.1 Navigation errors (1)
We identified several types of navigation errors. The
most frequently occurring of these was that the
2 http://www.class.ncdc.noaa.gov/saa/products/welcome
気象衛星センター技術報告 第 55 号 2011 年 2 月
– 3 –
positioning of some segments of the data stream seemed
odd. Fig. 1 shows the most obvious example of this type
of error. In a single file, such errors are usually found in
only a few scan lines (mostly 1 or 2 lines). However, the
number of files that contain this kind of navigation error
is quite large (this problem was encountered in more than
70% of files containing post-1995 data obtained from
DMSP-10).
Another type of navigation error that we found may
have been caused by the use of data from different
observation times. An example of this type of error is
shown in Fig. 2. This figure shows a composite of the
brightness temperatures for the SSM/I 37-GHz vertical
channel based on seven files of data obtained from
DMSP-13 on 29 December 2006. There is clearly a
discontinuity in the data obtained over the Pacific Ocean,
resulting in an unusually wide space in areas to the west
of this point. The reason for this inconsistency is that the
relevant data in
NPR.TDRR.S7.D06363.S0330.E0523.B6073031.NS
were from several days prior to the observation in
question. There are cases in which data from other
satellites are misinterpreted in this way, as well. Errors of
this kind are not common, but there were a few days for
which almost half the day’s data were erroneous.
We developed a method of detecting and eliminating
navigation errors by adopting a very simple navigation
technique. Fig. 3 shows brightness temperatures
processed by using the method. Almost all the erroneous
scan data have been eliminated. (This method does not
allow us to completely eliminate all errors, however, as
evidenced by the erroneous line that can be seen in data
for an area near the Antarctic.)
3.2 Detection of navigation errors (1)
The following is an explanation of the method used to
detect navigation errors. Because the orbit of a DMSP
satellite is almost circular, we assume the angular velocity
of a DMSP satellite to be constant. The position of the
satellite is given in latitude and longitude. Relative time
progress is measured from the time at which the satellite
Fig. 1 Brightness temperatures for the SSM/I 37-GHz vertical channel based on data obtained from DMSP-11
at 1611UTC on 25 March 1994 (NPR.TDRR.S5.D94084.S1611.E1739.A0011983.NS)
METEOROLOGICAL SATELLITE CENTER TECHNICAL NOTE, No.55, FEBRUARY, 2011
– 4 –
crosses the ascending node, and the longitude is measured
from the ascending node. First, we will explain the
navigation model and estimate the satellite position from
the footprints of the scan lines contained in a TDR file.
After that, we will describe the procedures used to check
for navigation errors.
Fig. 2 Composite of the brightness temperatures for the SSM/I 37-GHz vertical channel based on data obtained
from DMSP-13 on 29 December 2006 (NPR.TDRR.S7.D06363.S0330.E0523.B6073031.NS)
Fig. 3 Same as Fig. 1 but with erroneous data eliminated
気象衛星センター技術報告 第 55 号 2011 年 2 月
– 5 –
3.2.1 Prediction of the satellite’s position from the
navigation parameters
Let us assume that we know the satellite’s inclination,
the period of its rotation around the Earth (T), the time
that has elapsed since it crossed the ascending node, and
its ascending node (the time and the longitude). Using
these parameters, we can predict the future position of the
satellite.
Let A be the distance from the ascending node to the
satellite, B the latitude, and C the relative longitude of the
satellite, where γβα ,, are the angles shown in Fig. 4.
Notice that β =π - the inclination of the satellite orbit.
The distance A at the elapsed time TΔ from the node
can be expressed as follows.
TTA Δ×= /2π . (1)
Provided that 2/πα = , spherical trigonometry yields
the following equation, where the longitude and latitude
of the satellite at TΔ are:
.cos
cossinsin
,sinsinsin
BAC
ABβ
β×
=
×= (2)
Longitude C needs to be modified to take into
consideration the period of the Earth’s rotation (Te).
Because the Earth rotates by TTe Δ×/2π in TΔ , C
should be modified as follows.
TTeCC Δ×−=′ /2π
Given this, the satellite longitude C’’ is shown as:
''' 0 CCC −= (3)
We can now estimate the position of the satellite at
TΔ , provided we know the longitude of the node 0C
and the inclination.
3.2.2 Estimation of orbital parameters from the SSM/I
data stream
Orbital parameters such as β (inclination), and the
timing and longitude of the satellite’s ascending node can
be estimated by using the footprints in a scan line from
the SSM/I data stream.
First, we estimate the satellite’s position from the
footprints. Fig. 5 shows the relationship between the
satellite’s position (P, Q, and R) and the footprints in a
scan line (A, B, C) given that the SSM/I is a conical scan
sensor. While the SSM/I sensor scans the Earth from A to
Fig. 4 Illustration of a satellite orbit
METEOROLOGICAL SATELLITE CENTER TECHNICAL NOTE, No.55, FEBRUARY, 2011
– 6 –
C, the satellite moves from P to R and the shapes of the
footprints deviate slightly from being perfect semicircles.
However, we have been able to confirm that we can
approximate with sufficient precision that the footprints
make a circular arc from A, B to C with the center at Q, as
shown in the right hand side of Fig. 5 (if the direction
from P to R is constant, the two situations shown in Fig.5
lead to the same result).
The latitudes and longitudes of A, B, and C are
included in the TDR data stream, and L is a constant that
is known from the DMSP satellite specifications. By
using the spherical trigonometry shown in Appendix 1,
we can perform the following procedure.
(a) Calculate the distance between A and C and the
related angles (equations 1 and 2 in Appendix 1).
(b) Determine the latitude and longitude of M, which is
the central point between A and C (equation 3 in
Appendix 1).
(c) Determine the angle of NBQ (where N represents the
North or South Pole). We are able to determine this
because we know the latitudes and longitudes of B
and M (equation 2 in Appendix 1).
Determine the latitude and longitude of Q (the position
of the satellite). We are able to determine this because the
distance between B and Q is equal to L (equation 3 in
Appendix 1). We can also determine the direction from P
to B, which is assumed to be the direction of the velocity
of the satellite.
We now know the values for the following parameters
shown in Fig. 4: γ , B and 2/πα = . If we apply
spherical trigonometry again, we can determine the
inclinationβ :
γβ sincoscos B= . (4)
Once we know that, we can also determine the remaining
two parameters shown in Fig. 4, A and C, by means of
trigonometry, and then calculate the timing of the
ascending node by using equation (1) above.
With the exception of T (i.e. the period of the satellite’s
rotation around the Earth), all of the parameters that we
need to perform a navigation error compensation can be
estimated by using the procedure described above. We
can find out the nominal value of T (about 110 minutes)
from the specifications for DMSP satellites, but this value
is not precise enough for our purposes, and we found that
T changes over time. We can therefore obtain T from the
SSM/I data stream by monitoring the time at which the
Fig. 5 The relationship between a scan line and the movement of the satellite
気象衛星センター技術報告 第 55 号 2011 年 2 月
– 7 –
latitude of a scan line moved from the Southern
Hemisphere to the Northern Hemisphere in the
observations obtained from each orbit.
3.2.3 Correction of navigation errors (1)
The procedure for correcting navigation errors is as
follows.
1. Obtain the satellite position from a scan line of the
SSM/I data and calculate the orbital parameters from
the satellite position (see section 3.2.2).
2. Calculate the satellite position for the next scan line
by using the navigation model with the parameters
calculated in step 1 (see section 3.2.1).
3. Obtain the satellite position from the next scan line as
described in step 1 (see section 3.2.2).
4. Compare the satellite positions obtained in steps 2 and
3. If the difference between the positions is greater
than the specified criterion, the data for the second
scan line is discarded.
5. Repeat the above steps for the remaining scan lines.
When we calculate the satellite position and the orbital
parameters, we check whether the values obtained are
appropriate by comparing them with the nominal values
for the period of rotation T, the inclinationβ , and the
specified timing of the ascending node.
In step 4, criteria are empirically chosen and set for the
direction and speed of the satellite’s movement. The
precision of a satellite’s positioning is worse at higher
latitudes, and the precision with which a satellite’s
position can be predicted by using the orbital parameters
obtained based on this positioning is also worse at higher
latitudes. Therefore, the criteria set for higher latitudes
are greater than those set for lower latitudes. All scan
lines after the second one are checked according to the
above procedure, but the first scan line cannot be checked
in this way. Given this, if a QC of the second line results
in a fail, the first scan line is discarded instead of the
second one.
The type of navigation error shown in Fig. 1 can be
easily eliminated by following the above procedure, but
the type shown in Fig. 2 cannot be detected using this
procedure. The latter type of error can be detected by
using T, which is obtained from the TDR data. For the
purposes of our study, we assumed that the node is the
position located at the center of a scan data item when it
passes from the Southern Hemisphere into the Northern
Hemisphere, and so documented the longitude and time
for each orbit of the satellite. We then calculated T for
each orbit. The longitude of the node for the orbit under
observation was then compared with the longitude of the
node estimated by using the value for T obtained when
the node for the orbit was documented. If the difference
was greater than the specified criterion, the observation in
the file was discarded (in such cases, the data for the
whole day were discarded). However, the monitored T is
not precise enough to predict the ascending node beyond
a few days in advance. Consequently, in the computer
program we used, this method of performing a QC is only
efficient for up to three days.
3.3 Navigation errors (2)
An example of another kind of navigation error is
shown in Fig. 6. This figure shows brightness
temperatures in the 37-GHz vertical channel from DMSP-
08 at 2249UTC on 24 August 1987. Although the
presence of navigation errors is clearly visible in this
figure, no such errors were identified in the QC of the
data files. In microwave images, the brightness
temperatures of sea and land are quite distinct, so the
coasts of the continents are clearly visible in such images
and they should match a geographical map, as is the case
in Fig. 2, for example. However, the brightness
temperature distributions in Fig. 6 do not match the
geography, as the dark area in the brightness temperatures
around the east coast of North America should correspond
METEOROLOGICAL SATELLITE CENTER TECHNICAL NOTE, No.55, FEBRUARY, 2011
– 8 –
to the South American continent. We presume that the
positions and observation times recorded in the data file
are out of phase with the brightness temperature data.
This kind of error seems to occur only with pre-1994 data
files.
The method we adopted for dealing with this kind of
navigation data error is simple. We made use of the fact
that the brightness temperature of land is much higher
than that of an ocean. We counted the number of
observation points in the ocean area for which the
brightness temperature with a vertical polarization
component of 19 GHz was higher than the brightness
temperature criterion of 253 K. (This criterion is
considered to be the highest ocean temperature).
Therefore the data in which a brightness temperature in
the oceans exceeds the criterion should be discarded.
However, we often encountered data that were corrupted
by observation noise and geographical mapping errors.
After a few trials, we concluded that data should be
discarded if 10% of the observation points had higher
brightness temperatures than the criterion, on the basis
that this meant the data were navigationally erroneous.
This method works well in most cases, the exception
being cases where the number of ocean scans was very
small or the regions scanned were at high latitude because
the land–ocean temperature contrast is small at higher
latitudes. Given this, we only performed this QC for data
obtained from observation points located lower than 70
degrees latitude. This method will have to be further
refined in the future.
3.4 Format error encountered in converting the
original SSM/I data into netCDF format
The CLASS web site began offering SSM/I data in
netCDF format in August 2008 and, at present, SSM/I
TDR data before August 1997 are only available in
netCDF format. On finding out that there were some data
in netCDF format on CLASS that we did not have, we
downloaded the data and tried to make use of them.
Fig. 6 Brightness temperatures for the SSM/I 37-GHz vertical channel obtained based on data from DMSP-08
at 2249UTC on 24 August 1987 (NPR.TDRR.N8.D87236.S2249.E0032.B0093233.FN)
気象衛星センター技術報告 第 55 号 2011 年 2 月
– 9 –
However, we soon found that there was a problem with
the data. The data appeared to be valid if the observation
times attached to each scan line were ignored, but there
were some navigation errors like those found in the
original TDR data files. It appears that the SSM/I data in
netCDF format are not identical to the original TDR file
data. Instead of data having been extracted from every
scan line, they were extracted from every few scan lines
(i.e. certain data were culled), and it appears that in the
process of doing so, correspondence between the
observation time for each scan line and the relevant data
was lost. We need the precise observation time for each
scan line in order to be able to perform the navigation QC.
All of the netCDF files failed the QC we performed on
them, so we have not yet been able to use any SSM/I data
in the netCDF format. This problem needs to be resolved
in the next reanalysis project.
According to the communication that we received from
NCDC on 30 January 2010, this problem affects only
netCDF files containing data from between early 1993
and February 1997. NCDC are working to resolve this
problem (see the CLASS home page3 for further details).
4. Conclusion
The concluding remarks of this paper are as follows.
1. SSM/I data are very useful for obtaining information
regarding the surface of the Earth and its low-level
atmosphere. The long observation period over which
these data are collected makes them a particularly
attractive option.
2. However, the data contain some errors related to
navigation and brightness temperature. Although in most
cases such errors are found in just a few lines of a TDR
data file, we find a few TDR files in which all of the data
are rendered useless by navigation errors. Appendix 3
tabulates the TDR files of this kind.
3. We developed a method of mitigating navigation
errors, and adopted it for the preparation of observation
data for the JRA-55 reanalysis project. This method can
eliminate quite a few of the navigation errors found in
SSM/I data from DMSP satellites.
This method is not perfect, however, and we will have
to refine its performance or develop a better method. The
review and reprocessing of data on CLASS might prove
useful as a part of this development.
The source code for the computer program we
developed for navigation error compensation is given in
Appendix 2. The subroutine Chk_orbit is called each time
the data in an entire scan line is read out, and the program
then reports whether any navigation data errors have been
identified. Those interested are free to use this code,
either in part or in whole, without restriction. This
computer program is intended to be applied to a time-
ordered series of data files because it needs to monitor the
ascending node of the satellites. The efficiency and
reliability of the program is therefore limited. Persons
using this program will need to monitor the performance
manually, especially at the beginning of the process and
on days for which there are a lot of erroneous data.
Reference
Kazutoshi Onogi, 2007: The JRA-25 reanalysis, JMSC,
85, 369-432.
3 As of August 2010, the problem with the netCDF files on CLASS as described in this section has been completely resolved. The
correct version of the netCDF file is VER2_5. The authors are grateful for the NCDC’s prompt response to this issue and for their
efforts to confirm and resolve the problem.
METEOROLOGICAL SATELLITE CENTER TECHNICAL NOTE, No.55, FEBRUARY, 2011
– 10 –
Appendix 1: Some results obtained using spherical trigonometry
If we construct a triangle on the above globe with vertices A, B and O (where O represents the North Pole), we can
obtain the following formula.
)cos(coscossinsincos 122121 λλϕϕϕϕ −−=Θ (1)
βϕ
αϕ
θ sincos
sincos
sinsin 12 ==
Θ (2)
We can determine α and β from equation (2) if we know 11,λϕ and 22 ,λϕ , because we can determine
θsin/sinΘ from equation (1).
In addition, if we know Θ,, 11 λϕ , andα , then we can determine 2ϕ using equation (3), and because we know
θ from equation (2), we can then determine 2λ using equation (1).
αϕϕϕ cossincoscossinsin 112 Θ+Θ= (3)